Premup_analysis
This set of scripts are intended for the priors and flat versions of the PREMUP paradigm. We need to select below here which version to look at.
Compute individual performance indeces
# Initialize
strong_acc <- matrix(data=NA, nrow=40*4, ncol=max(exp.p$which_sub))
weak_acc <- matrix(data=NA, nrow=40*4, ncol=max(exp.p$which_sub))
flat_acc <- matrix(data=NA, nrow=40*4, ncol=max(exp.p$which_sub))
# temp <- NA
# Initialize plot
# par(mfrow=c(exp.p$nSubs/2,4), oma=c(0,0,0,0)) # Subplots
# Loop over subjects
for(cSub in exp.p$which_sub){
# Get subject code
sub_code <- paste("sub-", sprintf("%03d", cSub), sep="")
# Get one subject's data
sub_data <- subset(x = learning, subset = participant == cSub) # Data
load(paste(cleaned_dir, "../../data/BIDS/", sub_code,"/", sub_code, "_params.RData", sep="")) # Parameters
# Get cummulative accuracy split by scenes
for (cScn in seq(1,6, by=2)){
cumm_acc <- cumsum(sub_data$trial_acc_manual[sub_data$scn_cat==cScn])
cumm_acc2 <- cumsum(sub_data$trial_acc_manual[sub_data$scn_cat==cScn+1])
cumm_acc <- cumm_acc / c(1:length(cumm_acc)) # Cummulative accuracy
cumm_acc2 <- cumm_acc2 / c(1:length(cumm_acc2)) # Cummulative accuracy
# Get current scene's info
curr_scn_cond <- unique(sub_data$scn_condition[sub_data$scn_cat==cScn])
curr_scn_cond_label <- p$scn_cont[curr_scn_cond]
# Store average perfromance
if(curr_scn_cond_label==0.9){
strong_acc[1:length(cumm_acc),cSub] <- rowMeans(cbind(cumm_acc, cumm_acc2))
}else if(curr_scn_cond_label==0.7) {
weak_acc[1:length(cumm_acc),cSub] <- rowMeans(cbind(cumm_acc, cumm_acc2))
}else{
flat_acc[1:length(cumm_acc),cSub] <- rowMeans(cbind(cumm_acc, cumm_acc2))
}
# Plot
plot(cumm_acc, ylim = c(0,1), xlim = c(0,160),xaxt="n", col="pink",type="n",
main = paste(curr_scn_cond_label, "contingency"))
text(labels=sub_code, x=0, y=1.2, xpd=T)
abline(h=.5, col = "red", lty = 2); abline(h=curr_scn_cond_label ,col = "green")
abline(v=0, col = "black", lty = 4);abline(v=40, col = "black", lty = 4); abline(v=80, col = "black", lty = 4);
abline(v=120, col = "black", lty = 4);abline(v=160, col = "black", lty = 4);
xtick<-seq(0, 160, by=40)
axis(side=1, at=xtick, labels = FALSE)
text(x=xtick, par("usr")[3],
labels = c("1st", "2nd", "3rd", "4th", "hard-limit"), pos = 1, xpd = TRUE)
points(seq(1, length(cumm_acc2)), cumm_acc2, col = "blue")
}
}
Compute group level performance
# Overall performance
group_perf <- data.frame(
trial = c(1:160),
strong_scn = rowMeans(strong_acc, na.rm = T),
weak_scn = rowMeans(weak_acc, na.rm = T),
flat_scn = rowMeans(flat_acc, na.rm = T),
strong_sd = rowSds(strong_acc, na.rm = T),
weak_sd = rowSds(weak_acc, na.rm = T),
flat_sd = rowSds(flat_acc, na.rm = T)
)Plot
par(mfrow=c(1,3)) # Subplots
for(cCond in 1:3){
if(cCond == 1){ # Strong prior
plot(group_perf$strong_scn, ylim = c(0,1), xlim = c(0,160),xaxt="n", col="blue",
main = "Strong prior")
abline(h=max(p$scn_cont) ,col = "green")
# SD
xx <- group_perf$strong_scn+group_perf$strong_sd
yy <- group_perf$strong_scn-group_perf$strong_sd
} else if(cCond ==2){# Weak prior
plot(group_perf$weak_scn, ylim = c(0,1), xlim = c(0,160),xaxt="n", col="blue",
main = "Weak prior")
abline(h=p$scn_cont[2] ,col = "green")
# SD
xx <- group_perf$weak_scn+group_perf$weak_sd
yy <- group_perf$weak_scn-group_perf$weak_sd
} else {# Flat prior
plot(group_perf$flat_scn, ylim = c(0,1), xlim = c(0,160),xaxt="n", col="blue",
main = "Flat prior")
abline(h=min(p$scn_cont) ,col = "green")
# SD
xx <- group_perf$flat_scn+group_perf$flat_sd
yy <- group_perf$flat_scn-group_perf$flat_sd
}
# Format
abline(h=.5, col = "red", lty = 2);
abline(v=0, col = "black", lty = 4);abline(v=40, col = "black", lty = 4); abline(v=80, col = "black", lty = 4);
abline(v=120, col = "black", lty = 4);abline(v=160, col = "black", lty = 4);
xtick<-seq(0, 160, by=40)
axis(side=1, at=xtick, labels = FALSE)
text(x=xtick, par("usr")[3],
labels = c("1st", "2nd", "3rd", "4th", "hard-limit"), pos = 1, xpd = TRUE)
polygon(x= c(1:160, 160:1), y = c((xx),rev(yy)), col = rgb(.5,.5,.8,.3), border = "red")
}
# Last iteration performance
last_it <- learning[FALSE,]
last_strong_acc <- matrix(data=NA, nrow=40, ncol=max(exp.p$which_sub))
last_weak_acc <- matrix(data=NA, nrow=40, ncol=max(exp.p$which_sub))
last_flat_acc <- matrix(data=NA, nrow=40, ncol=max(exp.p$which_sub))
for(cSub in exp.p$which_sub){
sub_last_it <- with(learning, subset(x=learning,
subset =
participant == cSub &
iteration_index == exp.p$nIterations[exp.p$nIterations$participant==cSub,2]))
sub_code <- paste("sub-", sprintf("%03d", cSub), sep="")
load(paste(cleaned_dir, "../../data/BIDS/", sub_code,"/", sub_code, "_params.RData", sep=""))
# Store
last_it <- rbind(last_it, sub_last_it)
for (cScn in seq(1,6, by=2)){
cumm_acc <- cumsum(sub_last_it$trial_acc_manual[sub_last_it$scn_cat==cScn])
cumm_acc2 <- cumsum(sub_last_it$trial_acc_manual[sub_last_it$scn_cat==cScn+1])
cumm_acc <- cumm_acc / c(1:length(cumm_acc)) # Cummulative accuracy
cumm_acc2 <- cumm_acc2 / c(1:length(cumm_acc2)) # Cummulative accuracy
# Get current scene's info
curr_scn_cond <- unique(sub_last_it$scn_condition[sub_last_it$scn_cat==cScn])
curr_scn_cond_label <- p$scn_cont[curr_scn_cond]
# Store perfromance
if(curr_scn_cond_label==0.9){
last_strong_acc[1:length(cumm_acc),cSub] <- rowMeans(cbind(cumm_acc, cumm_acc2))
}else if(curr_scn_cond_label==0.7){
last_weak_acc[1:length(cumm_acc),cSub] <- rowMeans(cbind(cumm_acc, cumm_acc2))
}else {
last_flat_acc[1:length(cumm_acc),cSub] <- rowMeans(cbind(cumm_acc, cumm_acc2))
}
}
}
# Overall performance
group_perf_last <- data.frame(
trial = c(1:40),
strong_scn = rowMeans(last_strong_acc, na.rm = T),
weak_scn = rowMeans(last_weak_acc, na.rm = T),
flat_scn = rowMeans(last_flat_acc, na.rm = T),
strong_sd = rowSds(last_strong_acc, na.rm = T),
weak_sd = rowSds(last_weak_acc, na.rm = T),
flat_sd = rowSds(last_flat_acc, na.rm = T)
)Plot last iteration
par(mfrow=c(1,3), oma = c(0, 0, 0, 0)) # Subplots
for(cCond in 1:3){
if( cCond == 1){ # Strong prior
plot(group_perf_last$strong_scn, ylim = c(0,1), xlim = c(0,40),xaxt="n", col="blue",
main = "Strong prior", ylab="cumm-acc")
abline(h=max(p$scn_cont) ,col = "green")
# SD
xx <- group_perf_last$strong_scn+group_perf_last$strong_sd
yy <- group_perf_last$strong_scn-group_perf_last$strong_sd
} else if (cCond==2) {# Weak prior
plot(group_perf_last$weak_scn, ylim = c(0,1), xlim = c(0,40),xaxt="n", col="blue",
main = "Weak prior", ylab="cumm-acc")
abline(h=p$scn_cont[2] ,col = "green")
# SD
xx <- group_perf_last$weak_scn+group_perf_last$weak_sd
yy <- group_perf_last$weak_scn-group_perf_last$weak_sd
}else {# Flat prior
plot(group_perf_last$flat_scn, ylim = c(0,1), xlim = c(0,40),xaxt="n", col="blue",
main = "Flat prior")
abline(h=min(p$scn_cont) ,col = "green")
# SD
xx <- group_perf_last$flat_scn+group_perf_last$flat_sd
yy <- group_perf_last$flat_scn-group_perf_last$flat_sd
}
# Format
abline(h=.5, col = "red", lty = 2);
#xtick<-seq(0, 40, by=10)
#axis(side=1, at=xtick, labels = FALSE)
text(x=xtick, par("usr")[3],)
# labels = c("Last iteration"), pos = 1, xpd = TRUE)
polygon(x= c(1:40, 40:1), y = c((xx),rev(yy)), col = rgb(.5,.5,.8,.3), border = "red")
}
Save cleaned data
#write.csv(learning_data, file = paste(output_dir, sub_code, "_task-learning_cleaned.csv", sep=""))
#write.csv(rating_data, file = paste(output_dir, sub_code, "_task-rating_cleaned.csv", sep=""))